[PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jan Beulich]

From: Tim Deegan <Tim.Deegan@citrix.com>

Fixes a hang when booting as dom0 under Xen, when jiffies can be
quite large by the time the kernel init gets this far.

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
[jbeulich@novell.com: !time_after() -> time_before_eq() as suggested by Jiri Slaby]
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: stable@kernel.org 

---
 init/calibrate.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

--- 2.6.38-rc4/init/calibrate.c
+++ 2.6.38-rc4-calibrate-jiffy-overflow/init/calibrate.c
@@ -66,7 +66,7 @@ static unsigned long __cpuinit calibrate
 		pre_start = 0;
 		read_current_timer(&start);
 		start_jiffies = jiffies;
-		while (jiffies <= (start_jiffies + 1)) {
+		while (time_before_eq(jiffies, start_jiffies + 1)) {
 			pre_start = start;
 			read_current_timer(&start);
 		}
@@ -74,8 +74,8 @@ static unsigned long __cpuinit calibrate
 
 		pre_end = 0;
 		end = post_start;
-		while (jiffies <=
-		       (start_jiffies + 1 + DELAY_CALIBRATION_TICKS)) {
+		while (time_before_eq(jiffies, start_jiffies + 1 +
+					       DELAY_CALIBRATION_TICKS)) {
 			pre_end = end;
 			read_current_timer(&end);
 		}

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jan Beulich]

>>> On 09.02.11 at 21:54, Linus Torvalds <torvalds@linux-foundation.org> wrote:
> On Wed, Feb 9, 2011 at 12:31 AM, Jan Beulich <JBeulich@novell.com> wrote:
>> Fixes a hang when booting as dom0 under Xen, when jiffies can be
>> quite large by the time the kernel init gets this far.
>>
>> Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
>>
>> !time_after() -> time_before_eq() as per suggestion from Jiri Slaby.
>>
>> Signed-off-by: Jan Beulich <jbeulich@novell.com>
>> Cc: Jiri Slaby <jslaby@suse.cz>
>> Cc: Jeremy Fitzhardinge <jeremy@goop.org>
> 
> So I have nothing against the patch, but the changelog is broken.
> 
> Is Tim the original author? If so, why wasn't that noted? I'm just
> guessing from that random signed-off line in the middle of the thing.

Indeed, I forgot to add the From: line. I'm sorry about that; I'll
resubmit.

Jan

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Linus Torvalds]

On Wed, Feb 9, 2011 at 12:31 AM, Jan Beulich <JBeulich@novell.com> wrote:
> Fixes a hang when booting as dom0 under Xen, when jiffies can be
> quite large by the time the kernel init gets this far.
>
> Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
>
> !time_after() -> time_before_eq() as per suggestion from Jiri Slaby.
>
> Signed-off-by: Jan Beulich <jbeulich@novell.com>
> Cc: Jiri Slaby <jslaby@suse.cz>
> Cc: Jeremy Fitzhardinge <jeremy@goop.org>

So I have nothing against the patch, but the changelog is broken.

Is Tim the original author? If so, why wasn't that noted? I'm just
guessing from that random signed-off line in the middle of the thing.

Please guys, don't drop authorship attributions, or alternatively
don't confuse me with odd random sign-off lines in the middle of the
message.

                              Linus

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jeremy Fitzhardinge]

On 02/09/2011 05:18 AM, Jan Beulich wrote:
>>>> On 09.02.11 at 10:53, Jiri Slaby <jslaby@suse.cz> wrote:
>> On 02/09/2011 10:44 AM, Jan Beulich wrote:
>>>>>> On 09.02.11 at 10:21, Jiri Slaby <jslaby@suse.cz> wrote:
>>>> On 02/09/2011 09:31 AM, Jan Beulich wrote:
>>>>> Fixes a hang when booting as dom0 under Xen, when jiffies can be
>>>>> quite large by the time the kernel init gets this far.
>>>> As I wrote this might happen if the boot till this point takes ~ 5
>>>> minutes because we start at -5 minutes.
>>>>
>>>> That said, is this a candidate for stable? (If so, please CC stable.)
>>> Honestly, I'm not certain (given that mainline Xen Dom0 support
>>> is still only in its beginnings).
>> What about other VMs? If I run few non-kvm qemu VMs (kvm presets lpj) on
>> a busy dual-core machine, this will be a problem too, right?
> Yes, quite possible indeed.

Seems like stable material since it 1) has fixed a real bug, and 2) is
the right way to do things anyway.  It just hasn't been seen to fix a
bug in a stable kernel, so perhaps not.  If it were any more complex
then wouldn't push it.

    J

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jan Beulich]

>>> On 09.02.11 at 10:53, Jiri Slaby <jslaby@suse.cz> wrote:
> On 02/09/2011 10:44 AM, Jan Beulich wrote:
>>>>> On 09.02.11 at 10:21, Jiri Slaby <jslaby@suse.cz> wrote:
>>> On 02/09/2011 09:31 AM, Jan Beulich wrote:
>>>> Fixes a hang when booting as dom0 under Xen, when jiffies can be
>>>> quite large by the time the kernel init gets this far.
>>>
>>> As I wrote this might happen if the boot till this point takes ~ 5
>>> minutes because we start at -5 minutes.
>>>
>>> That said, is this a candidate for stable? (If so, please CC stable.)
>> 
>> Honestly, I'm not certain (given that mainline Xen Dom0 support
>> is still only in its beginnings).
> 
> What about other VMs? If I run few non-kvm qemu VMs (kvm presets lpj) on
> a busy dual-core machine, this will be a problem too, right?

Yes, quite possible indeed.

Jan

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jiri Slaby]

On 02/09/2011 10:44 AM, Jan Beulich wrote:
>>>> On 09.02.11 at 10:21, Jiri Slaby <jslaby@suse.cz> wrote:
>> On 02/09/2011 09:31 AM, Jan Beulich wrote:
>>> Fixes a hang when booting as dom0 under Xen, when jiffies can be
>>> quite large by the time the kernel init gets this far.
>>
>> As I wrote this might happen if the boot till this point takes ~ 5
>> minutes because we start at -5 minutes.
>>
>> That said, is this a candidate for stable? (If so, please CC stable.)
> 
> Honestly, I'm not certain (given that mainline Xen Dom0 support
> is still only in its beginnings).

What about other VMs? If I run few non-kvm qemu VMs (kvm presets lpj) on
a busy dual-core machine, this will be a problem too, right?

regards,
-- 
js
suse labs

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jan Beulich]

>>> On 09.02.11 at 10:21, Jiri Slaby <jslaby@suse.cz> wrote:
> On 02/09/2011 09:31 AM, Jan Beulich wrote:
>> Fixes a hang when booting as dom0 under Xen, when jiffies can be
>> quite large by the time the kernel init gets this far.
> 
> As I wrote this might happen if the boot till this point takes ~ 5
> minutes because we start at -5 minutes.
> 
> That said, is this a candidate for stable? (If so, please CC stable.)

Honestly, I'm not certain (given that mainline Xen Dom0 support
is still only in its beginnings).

Jan

Re: [PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jiri Slaby]

On 02/09/2011 09:31 AM, Jan Beulich wrote:
> Fixes a hang when booting as dom0 under Xen, when jiffies can be
> quite large by the time the kernel init gets this far.

As I wrote this might happen if the boot till this point takes ~ 5
minutes because we start at -5 minutes.

That said, is this a candidate for stable? (If so, please CC stable.)

> Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
> 
> !time_after() -> time_before_eq() as per suggestion from Jiri Slaby.
> 
> Signed-off-by: Jan Beulich <jbeulich@novell.com>
> Cc: Jiri Slaby <jslaby@suse.cz>
> Cc: Jeremy Fitzhardinge <jeremy@goop.org>
> 
> ---
>  init/calibrate.c |    6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> --- 2.6.38-rc4/init/calibrate.c
> +++ 2.6.38-rc4-calibrate-jiffy-overflow/init/calibrate.c
> @@ -66,7 +66,7 @@ static unsigned long __cpuinit calibrate
>  		pre_start = 0;
>  		read_current_timer(&start);
>  		start_jiffies = jiffies;
> -		while (jiffies <= (start_jiffies + 1)) {
> +		while (time_before_eq(jiffies, start_jiffies + 1)) {
>  			pre_start = start;
>  			read_current_timer(&start);
>  		}
> @@ -74,8 +74,8 @@ static unsigned long __cpuinit calibrate
>  
>  		pre_end = 0;
>  		end = post_start;
> -		while (jiffies <=
> -		       (start_jiffies + 1 + DELAY_CALIBRATION_TICKS)) {
> +		while (time_before_eq(jiffies, start_jiffies + 1 +
> +					       DELAY_CALIBRATION_TICKS)) {
>  			pre_end = end;
>  			read_current_timer(&end);
>  		}

thanks,
-- 
js
suse labs

[PATCH] fix jiffy calculations in calibrate_delay_direct to handle overflow

[Jan Beulich]

Fixes a hang when booting as dom0 under Xen, when jiffies can be
quite large by the time the kernel init gets this far.

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>

!time_after() -> time_before_eq() as per suggestion from Jiri Slaby.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>

---
 init/calibrate.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

--- 2.6.38-rc4/init/calibrate.c
+++ 2.6.38-rc4-calibrate-jiffy-overflow/init/calibrate.c
@@ -66,7 +66,7 @@ static unsigned long __cpuinit calibrate
 		pre_start = 0;
 		read_current_timer(&start);
 		start_jiffies = jiffies;
-		while (jiffies <= (start_jiffies + 1)) {
+		while (time_before_eq(jiffies, start_jiffies + 1)) {
 			pre_start = start;
 			read_current_timer(&start);
 		}
@@ -74,8 +74,8 @@ static unsigned long __cpuinit calibrate
 
 		pre_end = 0;
 		end = post_start;
-		while (jiffies <=
-		       (start_jiffies + 1 + DELAY_CALIBRATION_TICKS)) {
+		while (time_before_eq(jiffies, start_jiffies + 1 +
+					       DELAY_CALIBRATION_TICKS)) {
 			pre_end = end;
 			read_current_timer(&end);
 		}